kernel/rcupdate.c

   1 /*
   2  * Read-Copy Update mechanism for mutual exclusion
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License as published by
   6  * the Free Software Foundation; either version 2 of the License, or
   7  * (at your option) any later version.
   8  *
   9  * This program is distributed in the hope that it will be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write to the Free Software
  16  * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
  17  *
  18  * Copyright IBM Corporation, 2001
  19  *
  20  * Authors: Dipankar Sarma <dipankar@in.ibm.com>
  21  *          Manfred Spraul <manfred@colorfullife.com>
  22  *
  23  * Based on the original work by Paul McKenney <paulmck@us.ibm.com>
  24  * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen.
  25  * Papers:
  26  * http://www.rdrop.com/users/paulmck/paper/rclockpdcsproof.pdf
  27  * http://lse.sourceforge.net/locking/rclock_OLS.2001.05.01c.sc.pdf (OLS2001)
  28  *
  29  * For detailed explanation of Read-Copy Update mechanism see -
  30  *              http://lse.sourceforge.net/locking/rcupdate.html
  31  *
  32  */
  33 #include <linux/types.h>
  34 #include <linux/kernel.h>
  35 #include <linux/init.h>
  36 #include <linux/spinlock.h>
  37 #include <linux/smp.h>
  38 #include <linux/interrupt.h>
  39 #include <linux/sched.h>
  40 #include <linux/atomic.h>
  41 #include <linux/bitops.h>
  42 #include <linux/percpu.h>
  43 #include <linux/notifier.h>
  44 #include <linux/cpu.h>
  45 #include <linux/mutex.h>
  46 #include <linux/export.h>
  47 #include <linux/hardirq.h>
  48 #include <linux/delay.h>
  49
  50 #define CREATE_TRACE_POINTS
  51 #include <trace/events/rcu.h>
  52
  53 #include "rcu.h"
  54
  55 #ifdef CONFIG_PREEMPT_RCU
  56
  57 /*
  58  * Preemptible RCU implementation for rcu_read_lock().
  59  * Just increment ->rcu_read_lock_nesting, shared state will be updated
  60  * if we block.
  61  */
  62 void __rcu_read_lock(void)
  63 {
  64         current->rcu_read_lock_nesting++;
  65         barrier();  /* critical section after entry code. */
  66 }
  67 EXPORT_SYMBOL_GPL(__rcu_read_lock);
  68
  69 /*
  70  * Preemptible RCU implementation for rcu_read_unlock().
  71  * Decrement ->rcu_read_lock_nesting.  If the result is zero (outermost
  72  * rcu_read_unlock()) and ->rcu_read_unlock_special is non-zero, then
  73  * invoke rcu_read_unlock_special() to clean up after a context switch
  74  * in an RCU read-side critical section and other special cases.
  75  */
  76 void __rcu_read_unlock(void)
  77 {
  78         struct task_struct *t = current;
  79
  80         if (t->rcu_read_lock_nesting != 1) {
  81                 --t->rcu_read_lock_nesting;
  82         } else {
  83                 barrier();  /* critical section before exit code. */
  84                 t->rcu_read_lock_nesting = INT_MIN;
  85 #ifdef CONFIG_PROVE_RCU_DELAY
  86                 udelay(10); /* Make preemption more probable. */
  87 #endif /* #ifdef CONFIG_PROVE_RCU_DELAY */
  88                 barrier();  /* assign before ->rcu_read_unlock_special load */
  89                 if (unlikely(ACCESS_ONCE(t->rcu_read_unlock_special)))
  90                         rcu_read_unlock_special(t);
  91                 barrier();  /* ->rcu_read_unlock_special load before assign */
  92                 t->rcu_read_lock_nesting = 0;
  93         }
  94 #ifdef CONFIG_PROVE_LOCKING
  95         {
  96                 int rrln = ACCESS_ONCE(t->rcu_read_lock_nesting);
  97
  98                 WARN_ON_ONCE(rrln < 0 && rrln > INT_MIN / 2);
  99         }
 100 #endif /* #ifdef CONFIG_PROVE_LOCKING */
 101 }
 102 EXPORT_SYMBOL_GPL(__rcu_read_unlock);
 103
 104 /*
 105  * Check for a task exiting while in a preemptible-RCU read-side
 106  * critical section, clean up if so.  No need to issue warnings,
 107  * as debug_check_no_locks_held() already does this if lockdep
 108  * is enabled.
 109  */
 110 void exit_rcu(void)
 111 {
 112         struct task_struct *t = current;
 113
 114         if (likely(list_empty(&current->rcu_node_entry)))
 115                 return;
 116         t->rcu_read_lock_nesting = 1;
 117         barrier();
 118         t->rcu_read_unlock_special = RCU_READ_UNLOCK_BLOCKED;
 119         __rcu_read_unlock();
 120 }
 121
 122 #else /* #ifdef CONFIG_PREEMPT_RCU */
 123
 124 void exit_rcu(void)
 125 {
 126 }
 127
 128 #endif /* #else #ifdef CONFIG_PREEMPT_RCU */
 129
 130 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 131 static struct lock_class_key rcu_lock_key;
 132 struct lockdep_map rcu_lock_map =
 133         STATIC_LOCKDEP_MAP_INIT("rcu_read_lock", &rcu_lock_key);
 134 EXPORT_SYMBOL_GPL(rcu_lock_map);
 135
 136 static struct lock_class_key rcu_bh_lock_key;
 137 struct lockdep_map rcu_bh_lock_map =
 138         STATIC_LOCKDEP_MAP_INIT("rcu_read_lock_bh", &rcu_bh_lock_key);
 139 EXPORT_SYMBOL_GPL(rcu_bh_lock_map);
 140
 141 static struct lock_class_key rcu_sched_lock_key;
 142 struct lockdep_map rcu_sched_lock_map =
 143         STATIC_LOCKDEP_MAP_INIT("rcu_read_lock_sched", &rcu_sched_lock_key);
 144 EXPORT_SYMBOL_GPL(rcu_sched_lock_map);
 145 #endif
 146
 147 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 148
 149 int debug_lockdep_rcu_enabled(void)
 150 {
 151         return rcu_scheduler_active && debug_locks &&
 152                current->lockdep_recursion == 0;
 153 }
 154 EXPORT_SYMBOL_GPL(debug_lockdep_rcu_enabled);
 155
 156 /**
 157  * rcu_read_lock_bh_held() - might we be in RCU-bh read-side critical section?
 158  *
 159  * Check for bottom half being disabled, which covers both the
 160  * CONFIG_PROVE_RCU and not cases.  Note that if someone uses
 161  * rcu_read_lock_bh(), but then later enables BH, lockdep (if enabled)
 162  * will show the situation.  This is useful for debug checks in functions
 163  * that require that they be called within an RCU read-side critical
 164  * section.
 165  *
 166  * Check debug_lockdep_rcu_enabled() to prevent false positives during boot.
 167  *
 168  * Note that rcu_read_lock() is disallowed if the CPU is either idle or
 169  * offline from an RCU perspective, so check for those as well.
 170  */
 171 int rcu_read_lock_bh_held(void)
 172 {
 173         if (!debug_lockdep_rcu_enabled())
 174                 return 1;
 175         if (rcu_is_cpu_idle())
 176                 return 0;
 177         if (!rcu_lockdep_current_cpu_online())
 178                 return 0;
 179         return in_softirq() || irqs_disabled();
 180 }
 181 EXPORT_SYMBOL_GPL(rcu_read_lock_bh_held);
 182
 183 #endif /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
 184
 185 struct rcu_synchronize {
 186         struct rcu_head head;
 187         struct completion completion;
 188 };
 189
 190 /*
 191  * Awaken the corresponding synchronize_rcu() instance now that a
 192  * grace period has elapsed.
 193  */
 194 static void wakeme_after_rcu(struct rcu_head  *head)
 195 {
 196         struct rcu_synchronize *rcu;
 197
 198         rcu = container_of(head, struct rcu_synchronize, head);
 199         complete(&rcu->completion);
 200 }
 201
 202 void wait_rcu_gp(call_rcu_func_t crf)
 203 {
 204         struct rcu_synchronize rcu;
 205
 206         init_rcu_head_on_stack(&rcu.head);
 207         init_completion(&rcu.completion);
 208         /* Will wake me after RCU finished. */
 209         crf(&rcu.head, wakeme_after_rcu);
 210         /* Wait for it. */
 211         wait_for_completion(&rcu.completion);
 212         destroy_rcu_head_on_stack(&rcu.head);
 213 }
 214 EXPORT_SYMBOL_GPL(wait_rcu_gp);
 215
 216 #ifdef CONFIG_PROVE_RCU
 217 /*
 218  * wrapper function to avoid #include problems.
 219  */
 220 int rcu_my_thread_group_empty(void)
 221 {
 222         return thread_group_empty(current);
 223 }
 224 EXPORT_SYMBOL_GPL(rcu_my_thread_group_empty);
 225 #endif /* #ifdef CONFIG_PROVE_RCU */
 226
 227 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
 228 static inline void debug_init_rcu_head(struct rcu_head *head)
 229 {
 230         debug_object_init(head, &rcuhead_debug_descr);
 231 }
 232
 233 static inline void debug_rcu_head_free(struct rcu_head *head)
 234 {
 235         debug_object_free(head, &rcuhead_debug_descr);
 236 }
 237
 238 /*
 239  * fixup_init is called when:
 240  * - an active object is initialized
 241  */
 242 static int rcuhead_fixup_init(void *addr, enum debug_obj_state state)
 243 {
 244         struct rcu_head *head = addr;
 245
 246         switch (state) {
 247         case ODEBUG_STATE_ACTIVE:
 248                 /*
 249                  * Ensure that queued callbacks are all executed.
 250                  * If we detect that we are nested in a RCU read-side critical
 251                  * section, we should simply fail, otherwise we would deadlock.
 252                  * In !PREEMPT configurations, there is no way to tell if we are
 253                  * in a RCU read-side critical section or not, so we never
 254                  * attempt any fixup and just print a warning.
 255                  */
 256 #ifndef CONFIG_PREEMPT
 257                 WARN_ON_ONCE(1);
 258                 return 0;
 259 #endif
 260                 if (rcu_preempt_depth() != 0 || preempt_count() != 0 ||
 261                     irqs_disabled()) {
 262                         WARN_ON_ONCE(1);
 263                         return 0;
 264                 }
 265                 rcu_barrier();
 266                 rcu_barrier_sched();
 267                 rcu_barrier_bh();
 268                 debug_object_init(head, &rcuhead_debug_descr);
 269                 return 1;
 270         default:
 271                 return 0;
 272         }
 273 }
 274
 275 /*
 276  * fixup_activate is called when:
 277  * - an active object is activated
 278  * - an unknown object is activated (might be a statically initialized object)
 279  * Activation is performed internally by call_rcu().
 280  */
 281 static int rcuhead_fixup_activate(void *addr, enum debug_obj_state state)
 282 {
 283         struct rcu_head *head = addr;
 284
 285         switch (state) {
 286
 287         case ODEBUG_STATE_NOTAVAILABLE:
 288                 /*
 289                  * This is not really a fixup. We just make sure that it is
 290                  * tracked in the object tracker.
 291                  */
 292                 debug_object_init(head, &rcuhead_debug_descr);
 293                 debug_object_activate(head, &rcuhead_debug_descr);
 294                 return 0;
 295
 296         case ODEBUG_STATE_ACTIVE:
 297                 /*
 298                  * Ensure that queued callbacks are all executed.
 299                  * If we detect that we are nested in a RCU read-side critical
 300                  * section, we should simply fail, otherwise we would deadlock.
 301                  * In !PREEMPT configurations, there is no way to tell if we are
 302                  * in a RCU read-side critical section or not, so we never
 303                  * attempt any fixup and just print a warning.
 304                  */
 305 #ifndef CONFIG_PREEMPT
 306                 WARN_ON_ONCE(1);
 307                 return 0;
 308 #endif
 309                 if (rcu_preempt_depth() != 0 || preempt_count() != 0 ||
 310                     irqs_disabled()) {
 311                         WARN_ON_ONCE(1);
 312                         return 0;
 313                 }
 314                 rcu_barrier();
 315                 rcu_barrier_sched();
 316                 rcu_barrier_bh();
 317                 debug_object_activate(head, &rcuhead_debug_descr);
 318                 return 1;
 319         default:
 320                 return 0;
 321         }
 322 }
 323
 324 /*
 325  * fixup_free is called when:
 326  * - an active object is freed
 327  */
 328 static int rcuhead_fixup_free(void *addr, enum debug_obj_state state)
 329 {
 330         struct rcu_head *head = addr;
 331
 332         switch (state) {
 333         case ODEBUG_STATE_ACTIVE:
 334                 /*
 335                  * Ensure that queued callbacks are all executed.
 336                  * If we detect that we are nested in a RCU read-side critical
 337                  * section, we should simply fail, otherwise we would deadlock.
 338                  * In !PREEMPT configurations, there is no way to tell if we are
 339                  * in a RCU read-side critical section or not, so we never
 340                  * attempt any fixup and just print a warning.
 341                  */
 342 #ifndef CONFIG_PREEMPT
 343                 WARN_ON_ONCE(1);
 344                 return 0;
 345 #endif
 346                 if (rcu_preempt_depth() != 0 || preempt_count() != 0 ||
 347                     irqs_disabled()) {
 348                         WARN_ON_ONCE(1);
 349                         return 0;
 350                 }
 351                 rcu_barrier();
 352                 rcu_barrier_sched();
 353                 rcu_barrier_bh();
 354                 debug_object_free(head, &rcuhead_debug_descr);
 355                 return 1;
 356         default:
 357                 return 0;
 358         }
 359 }
 360
 361 /**
 362  * init_rcu_head_on_stack() - initialize on-stack rcu_head for debugobjects
 363  * @head: pointer to rcu_head structure to be initialized
 364  *
 365  * This function informs debugobjects of a new rcu_head structure that
 366  * has been allocated as an auto variable on the stack.  This function
 367  * is not required for rcu_head structures that are statically defined or
 368  * that are dynamically allocated on the heap.  This function has no
 369  * effect for !CONFIG_DEBUG_OBJECTS_RCU_HEAD kernel builds.
 370  */
 371 void init_rcu_head_on_stack(struct rcu_head *head)
 372 {
 373         debug_object_init_on_stack(head, &rcuhead_debug_descr);
 374 }
 375 EXPORT_SYMBOL_GPL(init_rcu_head_on_stack);
 376
 377 /**
 378  * destroy_rcu_head_on_stack() - destroy on-stack rcu_head for debugobjects
 379  * @head: pointer to rcu_head structure to be initialized
 380  *
 381  * This function informs debugobjects that an on-stack rcu_head structure
 382  * is about to go out of scope.  As with init_rcu_head_on_stack(), this
 383  * function is not required for rcu_head structures that are statically
 384  * defined or that are dynamically allocated on the heap.  Also as with
 385  * init_rcu_head_on_stack(), this function has no effect for
 386  * !CONFIG_DEBUG_OBJECTS_RCU_HEAD kernel builds.
 387  */
 388 void destroy_rcu_head_on_stack(struct rcu_head *head)
 389 {
 390         debug_object_free(head, &rcuhead_debug_descr);
 391 }
 392 EXPORT_SYMBOL_GPL(destroy_rcu_head_on_stack);
 393
 394 struct debug_obj_descr rcuhead_debug_descr = {
 395         .name = "rcu_head",
 396         .fixup_init = rcuhead_fixup_init,
 397         .fixup_activate = rcuhead_fixup_activate,
 398         .fixup_free = rcuhead_fixup_free,
 399 };
 400 EXPORT_SYMBOL_GPL(rcuhead_debug_descr);
 401 #endif /* #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD */
 402
 403 #if defined(CONFIG_TREE_RCU) || defined(CONFIG_TREE_PREEMPT_RCU) || defined(CONFIG_RCU_TRACE)
 404 void do_trace_rcu_torture_read(char *rcutorturename, struct rcu_head *rhp)
 405 {
 406         trace_rcu_torture_read(rcutorturename, rhp);
 407 }
 408 EXPORT_SYMBOL_GPL(do_trace_rcu_torture_read);
 409 #else
 410 #define do_trace_rcu_torture_read(rcutorturename, rhp) do { } while (0)
 411 #endif